Leapmotor's Second Brand Wants an iPhone-Style Revolution — Where Does Zhu Jiangming's Confidence Come From?

Leapmotor's Tech Day unveiled a second brand aiming for an iPhone-style replacement through spatial reconstruction, backed by full-domain in-house R&D and technology external supply.
At its Tech Day, Leapmotor proposed that its second brand pursue innovation like the iPhone replacing feature phones, centered on over 100% cabin space efficiency and vehicle architecture reconstruction. The article covers four dimensions: organization, space, intelligent driving, and technology external supply. Over ten years the company grew from 50 to 10,000 people, with 8,500 in R&D and 65% of vehicle cost self-developed and self-manufactured; the second brand reconstructs length, width, and height via an air-conditioning all-in-one unit, upright shock absorber layout, and flat low floor, while Clover 5.0 cuts the wiring harness to under 500 meters; intelligent driving shifts to the world model route, reducing code by 90% and reaching the top tier in under a year; technology external supply has empowered FAW and Stellantis, with electric drive and power electronics winning 8 OEM design wins. Zhu Jiangming stays restrained on humanoid robots, seeing them as only one or two years old, and focuses near-term on profitable automation scenarios.
At its Tech Day, Leapmotor put forward a bold enough self-positioning — not another product line, but a replacement akin to the iPhone displacing feature phones. The narrative core Zhu Jiangming set for the second brand is not price, but a reconstruction of spatial form and vehicle architecture. And with that comes the question: why does a carmaker known for cost-effectiveness believe it can pull off innovation of this magnitude?

From 50 People to 10,000: A Decade of an Engineer's Company
Leiphone learned that when Leapmotor was founded in 2015, the team had only 50 people, 41 of whom were engineers. Eleven years later, 21 of those 41 are still with the company. Today Leapmotor has 10,000 employees, 8,500 of whom do R&D; sales, HR, finance, and PR combined number only 1,500.
That ratio itself is a statement. Vehicle and electronics each have teams of roughly 3,000, and electric drive and battery each have over a thousand. The heads of all four product lines are veterans who joined in 2015. Zhu Jiangming sums up the company's model in two phrases: full-domain in-house R&D and deep self-manufacturing — set on day one, unchanged for ten years. The early-start electric drive, battery, and electrical/electronic architecture have already built a lead, and the later power supply and power electronics are all made in-house too.
Supporting this logic is a set of operating figures: 103,129 units delivered globally in August 2026, net profit attributable to shareholders of 210 million yuan in the first half of the year, 18 parts factories in hand, and 65% of vehicle cost under its own control. These numbers are the only hard evidence Zhu Jiangming can offer when he talks about "confidence."
Full-domain in-house R&D means a vehicle manufacturer does not rely on external suppliers and independently completes R&D across all core links from underlying hardware to upper-layer software, covering electric drive, battery, electrical/electronic architecture, smart cockpit, and intelligent driving. Deep self-manufacturing goes a step further by building its own factories on top of in-house R&D, converting research results directly into proprietary capacity. The combination means the company controls the bulk of vehicle cost and reduces bargaining dependence on the supply chain, but the price is massive upfront investment and high organizational complexity. Few carmakers in the industry truly implement this model; most brands only develop some modules in-house and procure the rest through Tier 1 suppliers. Leapmotor's claim that 65% of vehicle cost is under its own control is the direct financial expression of this model, and also the underlying chip that lets it dare to talk about technology external supply.
Over 100% Space Efficiency: A Reconstruction of Space
Zhu Jiangming's judgment is that new energy vehicles today cannot be distinguished from fuel vehicles on the outside; open the door and it's still five seats, six seats, seven seats — the functional zones have not changed at all. "This is like the old feature phones, which had just a few monotonous functions."
The engineering landing point of the second brand is to make the cabin counterintuitive. The metric Leapmotor gives is space efficiency over 100%, achieved by reconstructing along length, width, and height simultaneously, rather than squeezing space bit by bit within an established layout.
In length: the industry-first super-integrated air-conditioning all-in-one unit plus a rear-mounted powertrain — air-conditioning parts reduced by 18%, longitudinal space occupied reduced by 34%, piping reduced by 30%, and operating noise lowered by 3dB. The final cabin longitudinal length is about 4,400mm, roughly 700mm more than the same class, with interior length space up 19%. In width: an upright shock absorber layout with an aluminum die-cast integrated small tower package, with the shock absorber rotated outward 8.3 degrees, tucking back the two "horns" of traditional fuel vehicles, achieving a front-row transverse width of about 1,100mm and increasing leg-width space by 37.5%. In height: a flat low floor lowers the floor by about 70mm, and a native integrated smart lifting roof expands upward by about 800mm, increasing cabin height by 63.7%, allowing a person to stand upright inside — forming a Loft.

The thinking behind the air-conditioning all-in-one unit illustrates the point better than the specs. Zhu Jiangming draws an analogy with home air conditioners: the old window unit had few pipes, and one module realized all functions; but open the front compartment of a current car, and the compressor, condenser, evaporator plus a pile of piping are very scattered. "If we start from the end goal, we can thoroughly revolutionize the car's air conditioning."
At the electronic level it is also a rebuild. Clover 5.0 does true central integrated control: previously each domain controller made its own decisions; now only one central domain controller makes decisions, and other domain controllers only execute, compressing the vehicle wiring harness to under 500 meters. A 48V power supply throughout the vehicle builds an electric highway for the intelligent era, reducing harness weight by 50%; an asymmetric vehicle Ethernet scheme runs 10G downlink for large data flows and 100M uplink for small flows, with one cable carrying both video and control data simultaneously. Thermal management parts drop from 69 to 21, longitudinal space occupied is compressed from 580mm to 310mm, and piping shrinks from 6.2 meters to 3.3 meters; the industry-first 14-way valve connects cooling and heating circuits, supporting 48 switching modes and 54 energy interaction modes, with winter heating power of 18kW, 20% ahead of the industry.
According to the current plan, the second brand's new car built on this architecture will debut in Q4 next year and begin deliveries.
Space efficiency was originally a real estate term referring to the ratio of usable interior area to gross floor area. In the automotive context, it is usually borrowed to describe the ratio between effective passenger cabin space and the vehicle's external dimensions. Traditional fuel vehicles are limited by the layout of the engine, transmission, driveshaft, and exhaust system, so cabin space is often squeezed by mechanical structures and space efficiency is hard to raise. Electric vehicles eliminate the engine and driveshaft, but the air-conditioning assembly, battery pack thickness, and shock absorber towers still occupy large amounts of longitudinal and transverse space. Leapmotor's claim of over 100% space efficiency means that through the air-conditioning all-in-one unit, rear-mounted powertrain, upright shock absorber layout, and flat low floor, the actual usable space inside the cabin exceeds the baseline calculated by conventional proportions — essentially returning mechanically occupied space to the occupants.
Intelligent Driving: From Absent to Top Tier in Under a Year
In the past few years, if anyone made a list of China's top-tier intelligent driving players, Leapmotor would most likely not be on it. The company has consistently adhered to full-domain in-house R&D while controlling overall resource investment, from self-developed perception hardware and self-developed intelligent driving chips to self-developed software algorithms — at the cost of long being absent from mainstream discussion.
The change happened in the past year. In September 2025, Leapmotor committed to the world model route; in March 2026 it got a DEMO version of world-model intelligent driving running; and now the real vehicle has already been tested end-to-end by media and users in advance, reaching the industry's top tier. From choosing the route to standing up there took less than a year.
In route selection, Leapmotor does not rely on BEV perception or rule-based code, using LiDAR only as safety redundancy. On the cloud there is a foundation world model trained on massive data, giving the model the ability to understand the world and reason to generate the world; on the vehicle there is a real-time world model that reuses cloud capabilities and combines navigation with real-time vehicle state for real-time decisions, deployable on different compute levels such as 200TOPS, 640TOPS, and 1280TOPS. At the architecture level it was reconstructed from first principles: maintaining image-view representation, saving the compute of BEV conversion; designing a memory storage module so the model can remember history; and decoupling inertial and real-time understanding. Rules are used only to backstop traffic regulations such as traffic lights, so code volume drops by 90%.
Zhou Hongtao, head of Leapmotor's intelligent driving, disagrees with the "throw money and people at it" notion: "Most of our peers, for example those who previously invested thousands of people in intelligent driving, are now cutting back substantially. Often this is not something you can achieve just by throwing money and people at it." He likens his team to Sun Wukong, plucking a hair and turning it into a clone to do work.
The threshold has also been pressed very low: multiple compute platforms, and even cars under 100,000 yuan can carry it. As of the end of August, more than 300,000 vehicles equipped with LiDAR and high-end intelligent driving chips had been delivered, and Leapmotor promises these users a free upgrade to the latest world model version and lifetime free use. Among them, the earliest ten-thousand-plus units, due to inconsistent hardware platforms, would require at least 50 million yuan for algorithm porting and reconstruction; Leapmotor decided to satisfy them anyway, on the grounds that these users bought their cars for high-end intelligent driving at the time.
World model is a technology route that has emerged in intelligent driving over the past two years. Its core idea is to let a model learn the operating laws of the world from large-scale data, thereby gaining the ability to understand, reason about, and generate physical environments, no longer relying on hand-written rule code or explicit BEV spatial conversion. The cloud foundation world model handles massive data training, while the vehicle-side real-time world model makes online decisions combining navigation and vehicle state, and can be deployed on different compute platforms. Compared with traditional BEV plus rule-code solutions, the advantage of the world model is stronger generalization and a greatly reduced code volume, but its requirements for data scale, training compute, and model compression and deployment are extremely high. By choosing this route, Leapmotor bypasses the industry's mainstream perception-plus-rules paradigm and starts directly from model capability, which is one reason it could iterate quickly in under a year.
Technology External Supply: The Real Quality of the Second Growth Curve
Another set of numbers at Tech Day is even more noteworthy: Leapmotor's technology has already begun to be sold to others. Vehicle-level technology empowerment for FAW and Stellantis Group; the electronics line has 5 external partnerships, battery systems 10, and electric drive and power electronics components have won design wins with 8 OEMs. Zhu Jiangming says technology external supply has become the company's second growth curve.
Why would peers buy? First, Leapmotor has really made the parts once through. Zhu Jiangming says that in in-house R&D and self-manufacturing, Leapmotor is very much like BYD: "We have made almost every electronic component," including the entire thermal management system, even making the compressor inside it ourselves. Second, there is a hard threshold: for core components developed and manufactured in-house, PPM and 3mis must be better than those of suppliers before they are allowed onto the market. Better quality is the premise, better cost comes second — the order cannot be reversed.
And the math works out. Leapmotor has calculated one case: if a partner chooses a packaged output of battery packs, electric drive, all controllers, and lights, compared with selecting a bunch of suppliers on its own, there is at least a 1,000 to 2,000 yuan cost advantage, while also bringing the benefits of shorter磨合 time and lower risk. There are two delivery methods: one is production in Leapmotor's own factories, a model already completed and mass-produced with FAW; the other is production in the partner's factory, with Leapmotor providing core components.
This capability comes from the organizational approach. Leapmotor's four product lines are independent internally, each responsible end-to-end from product planning, project approval, and development to delivery, with clear boundaries. But independence does not mean separation; during collaborative R&D they are one whole — a vehicle project forms one team, and electronics engineers are embedded into each project team.
Zhu Jiangming's understanding of the supply chain also explains this hybrid strategy. He boils China's supply chain choices down to learning from Apple or from Samsung: Apple has no factories of its own, but all materials and components for every order are specified by itself — the "ultimate model"; Samsung develops the entire industrial chain in-house and became global number one. Leapmotor chooses something between the two, partly self-manufactured and partly externally supplied. The reason is practical — automobiles are a low-margin industry, and 100% self-manufacturing means no way to shave peaks and fill valleys, with capacity fluctuations beyond its own control. "We need to absorb what others have advantages in and use it for ourselves, then I can build a greater company than his."
Geography is factored in too. Leapmotor concentrates its vehicle plants in two places in Zhejiang, and components with high transport costs must be located within 5 to 10 kilometers. This idea came from Zhu Jiangming's visit to Toyota's lean production before building cars: all of Toyota's supporting parts and vehicle plants in Japan are placed in Toyota City, with the shortest logistics, and during slack or peak seasons workers can be shifted over to balance output.
PPM stands for Parts Per Million, used in manufacturing to measure the number of defective products per million units; the lower the value, the more stable the quality. 3mis usually refers to the failure rate or early failure indicator within 3 months, reflecting reliability in the early stage of product delivery. In the auto parts industry, OEMs have strict admission standards for suppliers' PPM and 3mis; fail to meet them and you cannot enter the supply chain. Leapmotor applies this standard to itself, requiring that core components developed and manufactured in-house must be better than external suppliers on these two indicators before being allowed onto the market — essentially using supplier assessment logic to constrain its own manufacturing system. This is also the premise for technology external supply to be accepted by partners such as FAW and Stellantis — quality must pass first, and only then does cost advantage mean anything.
Restraint Amid Growth Anxiety
All these arrangements ultimately have to answer one question: where does growth come from. Zhu Jiangming says he thinks about this every day. This year the entire Chinese auto industry fell 20%, and new energy fell 6% from January to August — no growth, but negative growth.
Leapmotor's incremental growth this year comes from expanding its product line: the A series and D series cover everything from 65,800 yuan all the way to 300,000 yuan, and four of the ten models on sale this year were released this year, filling price bands it previously did not cover. The next growth point is already lined up: the second brand will debut and begin deliveries in Q4 next year; in overseas markets, because product launch pace is about two quarters slower than domestically, with the A10 just launched and the A05 not yet on sale, overseas volume ramp-up of new products will become another increment next year.
Toward hotter concepts, Leapmotor is instead restrained. It has already built a fully self-developed robot; the motor, controller, battery, perception algorithm, and computing platform it needs are things Leapmotor has already done in cars for ten years. But Zhu Jiangming compares today's robots to a person's age: "It is basically at the pre-research stage, at least it can run — that's only one or two years old. Maybe at five or six it can jog, but to work, you need to be 20, and there is still a lot of growing to do." Leapmotor will not make humanoids a focus in the near term, but will instead pay attention to good automation scenarios. As summed up at the launch event: a robot that can be taken for a stroll is no skill; a robot that can make money is presentable.
Back to the question in the title. Leapmotor's confidence does not come from any single hit model, but from a system built over ten years: making the parts once through, cutting the organization into four end-to-end product lines, and holding 65% of vehicle cost in its own hands. Whether the iPhone-style revolution can be delivered will only be known when the new car is delivered in Q4 next year, but the engineering chassis supporting this bet is indeed thicker than most people think.
One-line Summary
Leapmotor's Tech Day revealed that its second brand aims at an iPhone-style replacement through spatial reconstruction, while laying out its systemic chips in full-domain in-house R&D and technology external supply.
Paragraph Summary
At Tech Day, Leapmotor proposed that its second brand should pursue innovation akin to the iPhone replacing feature phones, with the core landing points being cabin space efficiency over 100% and vehicle architecture reconstruction. The article unfolds across four dimensions — organization, space, intelligent driving, and technology external supply: over ten years the company grew from 50 to 10,000 people, with 8,500 in R&D and 65% of vehicle cost self-developed and self-manufactured; the second brand achieves three-dimensional reconstruction in length, width, and height through an air-conditioning all-in-one unit, upright shock absorber layout, and flat low floor, while Clover 5.0 compresses the wiring harness to under 500 meters; intelligent driving shifts to the world model route, cutting code volume by 90%, and reaches the top tier in under a year; technology external supply has already empowered FAW and Stellantis, with electric drive and power electronics winning design wins from 8 OEMs. Zhu Jiangming remains restrained on humanoid robots, believing they are currently only at a one- or two-year-old stage, and in the near term is more focused on profitable automation scenarios.
Related articles

NVIDIA's 30,000 Developers Use Cursor, Code Commit Volume Triples
Cursor engineer Chris Diaz breaks down five organizational gaps blocking enterprise AI coding scale—strategy, risk tiering, governance, metrics, and operating model. NVIDIA's 30,000 developers use Cursor daily, tripling code commits while defect rates stay flat. The key: rewrite the SDLC, not add tools.

Ponytail: Giving AI Coding Agents a Senior Engineer's Mindset
Ponytail is a free GitHub tool that injects a senior engineer's mindset into AI coding agents, making them ask whether code is even necessary first. According to TamilDeck, it can compress 200 lines of AI-generated code to 30, cutting token costs, with support for Cursor rules and Claude Code skill integration—though security review still requires human oversight.

Complete Guide to Connecting Cursor AI to the GitHub MCP Server
A step-by-step guide to connecting Cursor AI to the GitHub MCP server: create a personal access token, configure mcp.json, store the token in the ENV block, and verify with a PR query.